A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation
Round 1
Reviewer 1 Report
Novelty:
The paper deals with filtering of 3D TLS Point cloud based on leaves data with use of manifold distance and normal estimation. The de-noising approach with use of manifold distance can be considered a novelty in the work.
Significance of Content:
I dont find the problem is significant, since there are available tools for denoising, For better judgment, especially available tools have to compared with the developped approach.
Quality of Presentation/Scientific Soundness
The method section would need some reshaping, since there are repeating terms in the headlines, that makes the content hard to follow.
Some additional remarks:
*** a detailed workflow schema is missing, since there are many repeating terms in the processing steps, it would be good to include a general overview before explanation of the method.
*** In the objectives, the authors give a list of the contributions, I would recommend to reshape the methods section according to this order.
***The details of Floyd algorithm can be removed, giving a reference would be enough.
*** For noise elimination, I would prefer to see also standard methods (e.g.SOR in CloudCompare) in the comparison
Author Response
Dear Dr/ Prof:
On behalf of my co-authors, we thank you very much for giving us an opportunity to revise our manuscript, we appreciate very much for your positive and constructive comments and suggestions on our manuscript entitled "A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation". We have studied reviewer’s comments carefully and have made multiple major revisions. We have tried our best to revise our manuscript according to your comments. The main corrections in the paper and the responds to the reviewer’s comments are as follows, which we would like to submit for your kind consideration.
Above all, we would like to express our great appreciation to reviewer for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Yours sincerely,
C H Hu. E-mail: [email protected]
Corresponding author: ZPan.
E-mail: 785122987@qq.com
Author Response File: 
Author Response.doc
Reviewer 2 Report
How do you decide the number of clusters in initial clustering of a leaf point cloud?
What algorithm is used for clustering and why?
How do you figure out the number of outlier clusters?
Please explain equation 6 (part be): Triangle Inequality: how distance between d(pi, pj) can be bigger than sum of distances d(pi, pk) and d(pk, pj)? Just provide a small example when this kind of situation arises.
Explain: lower and upper truncation are same for all clusters or they are different for each cluster?
For clustering you have to calculate a square distance matrix, using euclidean or any other distance type. Question is : How do you handle a large n-by-n distance matrix (Dnxn). This square matrix quickly become very large and issues of memory may arise. For example: if you have 200,000 points than distance matrix will a (200,000 x 200,000)/2 – 200000 = around 20 billion elements. To handle this large matrix, you may need a very large RAM on computing machine.
Figure 7: Please realign figure and text with it (slope, offset,..). Text are messing with figure. C change direction of y-label in both figures 7 and 8.
Author Response
Dear Dr/ Prof:
On behalf of my co-authors, we thank you very much for giving us an opportunity to revise our manuscript, we appreciate very much for your positive and constructive comments and suggestions on our manuscript entitled "A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation". We have studied reviewer’s comments carefully and have made multiple major revisions. We have tried our best to revise our manuscript according to your comments. The main corrections in the paper and the responds to the reviewer’s comments are as follows, which we would like to submit for your kind consideration.
Above all, we would like to express our great appreciation to reviewer for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Yours sincerely,
C H Hu. E-mail: [email protected]
Corresponding author: Z Pan.
E-mail: 785122987@qq.com
Author Response File: Author Response.doc
Reviewer 3 Report
Review of the article “A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation”
The study “A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation” aims at development of a new filtering methods of noise from point clouds. The objective is not clearly stated in the Objective sub-section and should be identified in one statement. I found the method and the presentation suitable for publication in the Remote Sensing journal, pending some improvements identified by the line. As a general note, I found the applicability limited to some specific applications, definitely not in current forestry. Also, there are many missing descriptive statistics that would provide a more comprehensive understanding of the data. Images are not enough. Having said that, leaves with so many points, more than 200/leaf are not achievable on routine TLS. So I fail to see if this study can be applied to many leaves from the same tree. Furthermore, it seems that the leaves that received so many impulses are on the lower part of the crown, which are the least important for the tree. Therefore, the reader would benefit significantly from a more complete description of the data.
Also, there is no real Discussion chapter, just a kind of Introduction. A proper discussion should be developed, with why the proposed method is better than existing one. What are the pitfalls and limitations of the new method? I think that in the Discussion could be addressed the relationship between the method and the data, as the number of points.
Below are my detailed comments:
L12-13: “as the most important index in leaf morphology” - leaf area is definitely not THE most important index. it is ONE of the most important. Please adjust.
L15-16: “These outliers can be outlier clusters or sparse outlier points”- three times "outliers". Please rephrase as awkward.
L41: “hot”- unprofessional term, please replace.
L42: “technology”- eliminate as redundant
L54: “significant” - in-proper word. Could be "needed" instead of "significant"?
L56: replace "cause" with "source".
L57: eliminate “However” as not related with the previous sentences directly.
L62-63: “Salt-and-pepper noise is a noise caused by signal pulse intensity, which leads to black and white noisy points randomly appearing on the image”-salt and paper is a term that refers to images, such as radar images, but not to 3D data. the connection to lidar is weak and I suggest rephrase or elimination
L67: “surface model.” - which mode? there was no reference o a model.
L69: “adaptation”- word repetition, please replace.
L84-85: “introduced to achieve state-of-the-art performance and robustness to high levels of noise” - this is just blurb, as all algorithms were at a certain point "state of the art".
L92: because it follows and enumeration the period after “noise” should a colon.
L113: “modeling”- scanner do not produce model they record returns. please eliminate.
L115: “360 -270 .”- the angles refers to what. Provide the rest of the information: 360 horizontal and 270 vertical
L116: “differential lateral”- replace with "different"
L116-117: “which was located in the center of the experimental target.”- eliminate, as the target was obvious the tree.
L117: “procedure was finished,”- eliminate, as redundant.
L120: “many individual trees,”- replace with "several species"
L154: “In data analysis,”- eliminate, as irrelevant
L154: “statistical method”- replace with "technique"
L155:” It was run throughout the whole process of the proposed method in this study.”- eliminate, as redundant.
L186-188: “The shortest path problem, as a typical problem in graph theory, has been used in many fields. Since it emerged in the early 20th century and became active in the 1990s, many scholars have studied this issue [37].”- eliminate, as irrelevant
L201-202: “Floyd, Dijkstra, Bellman-Ford, and SPFA algorithms.”- provide reference for these algorithms.
L292: “clusters:”- eliminate, as just said.
L292-293: “points without outlier points”- this makes no sense. Please rephrase.
L304-306: “Point cloud data for experimental hardwood tree groups were collected using a Leica C10 on the campus of Nanjing Forestry University, and then many well isolated”- eliminate, as stated in the Methods already.
L321: “Real leaf point cloud.”- this is a huge number of points for one leaf. For a tree, the density will be in many hundreds of millions. I really doubt that this data can be meaningful in real applications. More details on data are needed; and I mean really a lot more, including the time spent at each location, how many locations for the TLS, distance to the trees and how the points from each locations were merged.
L327: “reveals good filtering performance”- how "good" is measured? a significant amount of points that are inside the leaf are identified as outliers? What happens if there are just few points not that many?
L334-335: “For each leaf, the filtering performance, point cloud after filtering and 3D reconstruction results for the three methods are given in the corresponding figures”- eliminate, as this is a caption of a figure.
L351: “more smooth edges”- how is this measured? please provide measures for subjective statements, such as big small, good, smooth etc.
L414: “large”- please be specific about qualitative statements: how large is large?
L418: “60 leaf samples”-First, why 60 leaves and not 1000 or 6000, given the number of trees and the number of leaves/tree. This seems as only few leaves have many points to allows the algorithm to perform as expected. Second, provides the summary statistics for the leaves (length, width, area with the corresponding mean and variance, not only the range) and of the point clouds of each leaf (total number/leaf, average density /leaf)
L425 Fig. 7: “Offset:”- what offset is? There is no such a thing in regression analysis. Is this the intercept? where are the p -value for each coefficient.
L442: “n = 40: - why 40 here and 60 before? what happened with the rest of the 20?
L442: : The comparison results are reported in Table 1.”- eliminate as not needed.
L450-473: this belongs in Introduction, not discussions.
L473-475: “This study used the manifold distance to construct a local plane surface and determine which points should contribute to the normal estimations.”- eliminate, as said in the Methods.
L475-476: “Figures 4,5,6 show that using the manifold distance can achieve good filtering performance compared with the traditional Euclidean distance and classical PCA.”- eliminate, as said in the Results.
L480: “satisfactory.”- what satisfactory means ? how it was assessed.
L487-489:” The experimental results in Section 3.2 show that filtering the point clouds using the proposed method resulted in better reconstruction and visualization performance”- eliminate, as said in the results.
L502: “experiments”- replace experiment with "analyses". Do this for the subsequent mentions.
Author Response
Dear Dr/ Prof:
On behalf of my co-authors, we thank you very much for giving us an opportunity to revise our manuscript, we appreciate very much for your positive and constructive comments and suggestions on our manuscript entitled "A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation". We have studied reviewer’s comments carefully and have made multiple major revisions. We have tried our best to revise our manuscript according to your comments. The main corrections in the paper and the responds to the reviewer’s comments are as follows, which we would like to submit for your kind consideration.
Above all, we would like to express our great appreciation to reviewer for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Yours sincerely,
C H Hu. E-mail: [email protected]
Corresponding author: Z Pan.
E-mail: [email protected]
Author Response File: Author Response.doc
Round 2
Reviewer 1 Report
The comments, suggestions from the authors are well addressed.
Author Response
Dear Dr/ Prof:
We thank you very much for your response. Your affirmation on our first revision of the manuscript entitled "A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation" is indeed an incentive for us. We appreciate very much for your careful review and giving us positive and constructive comments and suggestions in the previous manuscript.
Thank you very much for all your help.
Best regards.
Yours sincerely,
C H Hu. E-mail: [email protected]
Corresponding author: Zhou Pan. E-mail: 785122987@qq.com
Author Response File: Author Response.doc
Reviewer 3 Report
The paper is an improvement over the previous version but the narrative is marred by many comparative words, which have no place in a scientific discourse, such as "novel" or "great inferiority". If the authors developed something, that will be automatically novel,so no need of duplication. If they mirror some old work that is called "replicate". Please review the manuscript for all comparisons and address them.
I am still not convinced about the applicability of the method in real situations. Fig. 4 clearly shows that there are many points per leaf, definitively more than 200 as the authors stated in the rebuttal letter, which will make scanning very long and very hard to manipulate the point cloud. Taking the question one level more, if the leaves, which are moving objects, have so many points, how many points the stem and branches get?
Also, the authors does not provide the meaningful statistics about the point cloud (points/tree, points/sq.m (even that is not that relevant in this case but provides crown size information),total height of the tree or dbh, to have an idea of the size of the trees, etc). This should be included.
Because the manuscript does not have line numbers, which is annoying to say the least, I am making just one comments, in the sense that some words are missing throughout the manuscript, such as the "is" before "superior" in the Discussions: "Hence, a new one that "missing is" superior to many". Here is another point "one" what", method? If so, please state method, rather than let the reader infer which "one" the authors refers too.
Author Response
Dear Dr/ Prof:
We thank you very much for your affirmation on our first revision of the manuscript entitled "A 3D Point Cloud Filtering Method for Leaves Based on Manifold Distance and Normal Estimation" and we appreciate very much for giving us another opportunity to revise it again. We have tried our best to study the comments and have made multiple revisions according to them. The main corrections in the paper and the responds to the comments are as follows, which we would like to submit for your kind consideration.
Above all, we would like to express our great appreciation to your comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Yours sincerely,
C H Hu. E-mail: [email protected]
Corresponding author: Zhou Pan. E-mail: 785122987@qq.com
Author Response File: Author Response.doc
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
